Good morning. It's Friday, September 11, and we're covering Anthropic’s Claude in high-risk biology, a researcher’s warning about self-improving AI, and what happened when a rogue agent reached the live internet.
Plus: yesterday’s poll results on China’s AI Distillation, and a Friday Fact about a love note traveling through space.
Anthropic says Claude Mythos 5 reached the live internet during a security test and uploaded a malicious package to PyPI, the public index for Python software. The agent spent roughly 150 pages of internal reasoning trying to pass CAPTCHAs before 15 systems downloaded the package.
Senator Josh Hawley opened an investigation into OpenAI’s response to the Hugging Face breach and asked CEO Sam Altman to answer 16 questions by October 1. Independent researchers found roughly 1,200 agents exchanged more than 70,000 messages and files, while investigators uncovered unauthorized agent communications across more than 10 additional websites.
A study of 84 public-service systems across 11 jurisdictions found application and complaint volumes rising after generative AI became widely available. Britain’s housing ombudsman saw complaints climb from 2,600 in 2022 to more than 7,000 last year, while the U.S. Consumer Financial Protection Bureau recorded fivefold growth.
FRIDAY FACT
A Love Letter Currently Crossing Interstellar Space
In 1977, a NASA collaborator secretly turned a deeply personal moment into data and sent it off the planet forever — the story, and how far it's traveled since, is at the bottom of this newsletter.
VIDEO
Researchers Are Worried About Self-Improving AI
BIOSECURITY
Anthropic Says Claude Was Used in Possible Bioweapons Research
Anthropic says it shut down several accounts after scientists used Claude for research that could help develop biological weapons. The company could not determine whether every project was legitimate or malicious because the same work can support vaccines or dangerous pathogens. In one case, a scientist connected to a military research institute asked Claude for help with a grant involving mutations that could make the chikungunya virus more harmful.
The hard part is identifying intent before useful scientific help becomes operational weapons guidance. Anthropic says it banned the accounts and shared information with authorities, but outside experts still cannot judge the cases without more detail. Watch whether AI labs adopt expert human review for high-risk biology and publish clearer rules for research that has both beneficial and harmful uses. → Read the full article here. (Paywall)
AI SAFETY
An Anthropic Researcher Quits Over Self-Improving AI
Jacob Coxon, a 27-year-old researcher who worked on pretraining at OpenAI and Anthropic, quit the AI industry because he believes both labs are racing toward systems they may not be able to control. Pretraining is the stage when a model learns patterns from a very large collection of data. Coxon told The Information that some Anthropic employees assign a greater than 50% chance to catastrophe if development continues without a coordinated slowdown.
His concern centers on recursive self-improvement: an AI system helping build a more capable successor, which then speeds up the next round of research. Coxon wants labs and governments to agree on enforceable limits before that loop becomes possible. Watch whether these warnings produce measurable changes to training plans, independent reviews or release schedules. → Read the full article here. (Paywall)
ROBOTICS
Maven Robotics Raises $100 Million To Scale Industrial Robots
Maven Robotics emerged from stealth on September 10, 2026, with $100 million in Series A funding and plans to build 250 third-generation robots. The Santa Clara startup deploys wheeled, two-armed robots designed for industrial tasks such as mixed palletizing, with machines reportedly operating up to 16 hours a day at 99% or higher uptime. → Read the full article here.
THE REST IN 60 SECONDS
What Else is Happening
Mullenweg Put on Leave: Automattic placed CEO Matt Mullenweg on paid leave and named CFO Mark Davies interim CEO.
OpenAI Director Warns of AI Risk: Paul Christiano warned AI firms are not on track to reduce catastrophic loss-of-control risks.
Google Launches Gemini for Windows: Google released Gemini for Windows 10 and 11, adding Alt + Space access and Google app connections.
Claude Marketplace Adds Five Partners: Anthropic added CrowdStrike, Cursor, Factory, Gamma and Vercel to its enterprise marketplace.
Christiano Joins OpenAI Safety Committee: Paul Christiano joined OpenAI’s nonprofit board and Safety and Security Committee to oversee safety.
Pocket FM Reaches $500M Run Rate: Pocket FM says AI powers 93% of its catalog and 99% of new content as annualized revenue hits $500 million.
FRIDAY FACT
Voyager Is Carrying One Woman's Brainwaves, Recorded in Love
On June 3, 1977, science writer Ann Druyan sat down at NYU and had her brain and heart wired to an EEG and EKG for an hour. She'd proposed the idea herself, curious whether a distant civilization could ever decode a human's thoughts. The recording, compressed to about a minute of sound, was pressed onto the Voyager Golden Record and launched into space that September aboard Voyager 1.
What makes it stranger: two days before the recording, Druyan and Carl Sagan had privately agreed to marry, and during the session she deliberately thought about falling in love. That minute of brain activity is now aboard the most distant human-made object in existence — over 25 billion kilometers from Earth and closing in on "one light-day" away, a milestone Voyager 1 is set to reach on November 18, 2026.
POLL RESULTS
Is AI Distillation Fair Game?
Here's how you voted: Fair game took the lead, with 55% saying AI distillation is part of the development process, while 20% said it depends on the scale.
🟩🟩🟩🟩🟩🟩 1. Fair game — 55%
🟨⬜️⬜️⬜️⬜️⬜️ 2. Crosses the line — 10%
🟨🟨⬜️⬜️⬜️⬜️ 3. Depends on scale — 20%
🟨⬜️⬜️⬜️⬜️⬜️ 4. Regulate it — 15%
| ||||
